The present invention relates to an apparatus for detecting a fatigue level of a person and detecting whether the person is in a drowsy state and in a state of doze (hereinafter these states are both referred to as a “dozing state”) by analyzing voice, more specifically the person's speaking voice, and also relates to a recording medium on which a program for realizing such an apparatus with a general-purpose computer is recorded.
There have been instances of serious accidents caused by airplane pilots, air traffic controllers or drivers of public transport systems, such as trains, taxies, tracks and buses, who were dozing or in an extremely fatigued state, and there have also been many instances of various accidents, though they did not develop into serious accidents, in which such a doze or extremely fatigued state was a contributing factor. Conventionally, for example, whether a pilot, air traffic controller or driver is in such a state is judged by his/her superior's observation during the morning roll call, but the observation by human's eyes has its limitation and there is also a high possibility that his/her state will change thereafter.
In resent years, however, not only in the aviation field, but also in the public transport systems as mentioned above, it is possible to readily communicate with a driver by radio communication and to presume the driver's fatigue level to a certain extent from the tone of his/her voice during the communication. Moreover, there is such a system that it regards no communications over a certain period of time as a high possibility of occurrence of some abnormality, and provide sa warning. However, it is difficult in practice for a person to judge another person's fatigue level or judge accurately whether another person is in a dozing state by listening the speaking voice of another person.
In view of such circumstances, nowadays, for instance, there have been thoughts that the physical conditions, for example, the states of pulse and pupil, of a driver are detected with a sensor and transmitted for analysis by radio with the use of radio traffic facilities introduced in various public transport systems as mentioned above.
For example, Japanese Patent Application Laid-Open No. 9-308614 (1997) discloses a technique of knowing a person's mental and physical conditions from the pulse wave and electrocardiogram, and Japanese Patent Application Laid-Open No. 7-117595 (1995) discloses a technique of detecting a doze from the pulse wave. Further, in resent years, chaos analysis has been often used as a technique of analyzing the obtained data. For instance, Japanese Patent Application Laid-Open No. 4-208136 (1992) discloses a technique of detecting a person's health level, fatigue level and state of doze by chaos analysis of the person's pulse wave. In addition, Japanese Patent Application Laid-Open No. 7-116119 (1995) discloses a technique of judging a person's health level by chaos analysis of the area of pupil and the refractive index of the lens of the person's eye.
However, with the techniques based on the pulse wave, electrocardiogram, the area of pupil and the refractive index of the lens as mentioned above, it is necessary to attach some sensor to a person as a subject and to send the information obtained by the sensor by radio or cable, or to analyze the information on the spot by an analyzer. Moreover, in the case where a sensor is attached to a person's body, the person may fall into unusual mental and physical conditions due to the stress caused by the attachment of the sensor. In this case, the information about the person in normal conditions, which is to be used as criteria for judgment, can not be obtained, and thus there is a fear that accurate detection and judgment are not performed.
By the way, the inventors of this application calculated the Lyapunov exponents by chaos analysis of voices uttered by a person, compared the calculation results in a time series, and gained knowledge that the Lyapunov exponents are higher when the person's fatigue level is high or when the person falls into a doze. Accordingly, by calculating the Lyapunov exponents of voices uttered by a person at different points of time and comparing them, it is possible to detect the person's fatigue level and/or dozing state.